A month or year Shift of a short LocalDateRange can return Empty; R28 says Shift keeps the shape #259

Problem

#223's month rule (PR #238) shifts the exclusive end, so a short LocalDateRange ending on the 28th–31st can shift to Empty:

• [2024-02-28,2024-02-28].Shift(Period.FromYears(1)) is Empty: the start goes to 2025-02-28, and the end goes 02-29 → 2025-02-28 → minus one day → 02-27.
• [2024-03-30,2024-03-30].Shift(P1M), [2024-01-29].Shift(P1M) and [2023-12-30].Shift(P-1M) are Empty too.

A sweep of every day from 2023-12 to 2025-03, lengths 0–35, steps ±1M, ±1Y and 13M, found 33 such cases. PostgreSQL gives the same (daterange(d + interval, (d+1) + interval) is empty), and so does LocalDateTimeRange on [d,d+1). But claim R28 (docs/ranges.md#move-or-widen-a-range) says "Both keep the shape", and the docs mention Empty only for a negative Expand. #238's tests sweep whole months and quarters only. Found by the final pre-release analysis on add665c (pass A).

Decision (maintainer, 2026-10-19)

Keep the rule, since it is PostgreSQL's answer and LocalDateTimeRange's, and document it. A month or year Shift of a short range can return Empty.

Done when

R28 and R31 and docs/ranges.md say when Shift returns Empty. A test holds the sweep's representative cases ([2024-02-28,2024-02-28].Shift(P1Y) is Empty, and so on) for LocalDateRange, DateRange if it has period shifts, and LocalDateTimeRange.